Durham Research Online
Deposited in DRO:
10 January 2016
Version of attached �le:
Accepted Version
Peer-review status of attached �le:
Peer-reviewed
Citation for published item:
Beckett, C.T.S. and Hall, M.R. and Augarde, C.E. (2013) 'Macrostructural changes in compacted earthenconstruction materials under loading.', Acta geotechnica., 8 (4). pp. 423-438.
Further information on publisher's website:
http://dx.doi.org/10.1007/s11440-012-0203-6
Publisher's copyright statement:
The �nal publication is available at Springer via http://dx.doi.org/10.1007/s11440-012-0203-6
Additional information:
Use policy
The full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission or charge, forpersonal research or study, educational, or not-for-pro�t purposes provided that:
• a full bibliographic reference is made to the original source
• a link is made to the metadata record in DRO
• the full-text is not changed in any way
The full-text must not be sold in any format or medium without the formal permission of the copyright holders.
Please consult the full DRO policy for further details.
Durham University Library, Stockton Road, Durham DH1 3LY, United KingdomTel : +44 (0)191 334 3042 | Fax : +44 (0)191 334 2971
http://dro.dur.ac.uk
Macrostructural changes in compacted earthen construction1
materials under loading2
3
October 19, 20124
C.T.S. Beckett (corresponding author)5
School of Civil and Resource Engineering,6
University of Western Australia,7
Crawley, WA 60098
10
M.R. Hall11
Nottingham Centre for Geomechanics,12
Division of Materials, Mechanics and Structures,13
Faculty of Engineering,14
University of Nottingham,15
University Park,16
Nottingham, NG7 2RD, UK17
19
C.E. Augarde20
School of Engineering and Computing Sciences,21
Durham University,22
Durham, DH1 3LE, UK23
25
Keywords: X-ray computed tomography, rammed earth, macrostructure, cracking26
Abstract27
There is increasing interest in the use of earthen materials for modern construction. The28
mechanical behaviour of these materials is strongly controlled by their internal macrostruc-29
tures. Rammed earth (RE) is one example of these materials, created by in situ compaction30
of a wet soil mixture. Changes to the material structure occur on loading and during com-31
paction; therefore, the nature of these changes needs to be understood if the effect on the32
material behaviour can be predicted. Here, the change in the macrostructure of RE on the33
application of compressive loading is investigated by using X-ray computed tomography and34
fractal analysis to monitor the changes in loaded RE specimens. The macrostructures of35
specimens comprising different layer thicknesses are also investigated in order to determine36
how layer thickness affects the compaction of the material. Results are used to recommend37
procedures for manufacturing specimens that are representative of the material found in38
full-scale RE structures.39
1
1 Introduction40
Rammed earth (RE) is the name given to an ancient building technique and an earthen con-41
struction material. Until recently, RE has been regarded as a material for which structural42
design was based on the experience of local craftsmen, much as masonry was regarded in the43
early 20th century. However, the desire to use RE in new environments, encouraged by its44
inherent sustainability, low environmental impact and its ability to maintain comfortable living45
conditions, has prompted a new examination of the material as a compacted, highly unsaturated46
soil (Jaquin et al., 2009). In addition, the large number of heritage RE structures worldwide47
requiring conservation has prompted the search for better scientific understanding.48
RE’s behaviour is controlled by the internal structure of the material, namely the distribution49
of particles, pores and water and how these combine to form aggregates (Diamond, 1971; Collins50
and McGown, 1974; Tuller et al., 1999; Tarantino and De Col, 2008; Monroy et al., 2010; Zhang51
and Li, 2010; Beckett, 2011). Changes which occur to the internal structure therefore affect the52
behaviour of RE. Such changes can be due to cracking on the application of a load or due to the53
use of different compactive efforts in different regions of the material. It is therefore necessary54
to be able to predict how the material will change in order to know the subsequent effect on55
the material behaviour.56
The process of cracking in rocks has been studied in some detail, however much less attention57
has been paid to sedimentary materials, for example RE (Dehandschutter et al., 2004, 2005).58
An understanding of the cracking process in soils has gained importance over recent years due to59
concerns associated with the burial of nuclear waste, for example, where heat-induced fracturing60
can lead to changes of permeability of the lining material and the escape of potentially-dangerous61
leachates (Gens et al., 2011). For dry soils, the application of a tensile or compressive load results62
in brittle failure through the formation, growth and joining of flaws within the soil structure63
(Atkinson, 1987). These flaws can either be the pores themselves or pre-existing cracks. Brittle64
failure of specimens at low water contents is well documented and is due to frictional failure65
and the breaking of liquid bridges at the interparticle contacts (Braunack et al., 1979; Rondeau66
et al., 2003; Dehandschutter et al., 2005; Baltodano-Goulding, 2010; Hoyos et al., 2010; Vallejo,67
2010).68
An RE layer is formed by compacting loose sandy-loam subsoil at its optimum water con-69
tent (OWC) between formwork. However, due to friction with the sides of the formwork, the70
2
compactive effort applied to material towards the top of the layer is greater than that towards71
the base (Morel et al., 2007). This can be mitigated slightly by the use of specially-shaped72
rammer heads (e.g. heart– or wedge-shaped), but a change in the compactive effort is still73
present, so that the material structure, and so its behaviour, varies throughout the layer thick-74
ness (Proctor, 1933; Betts and Miller, 1937). As these changes in behaviour can be detrimental75
to the structure (e.g. reduced strength or increased permeability), it is necessary to understand76
how compaction of different layer thicknesses affects the material structure so that appropriate77
thicknesses can be used.78
Here, X-ray Computed Tomography (XRCT) is used to non-destructively investigate the79
change in the macrostructure of RE specimens on the application of a load. Specimens com-80
prising different layer thicknesses are also examined in order to determine how layer thickness81
affects the compaction of the material.82
2 X-ray computed tomography83
Computerised tomography (CT) was developed in order to create 2-D or 3-D reconstructions84
of internal features of objects non-destructively (Roscoe, 1970; Wellington and Vinegar, 1987).85
XRCT is an improvement over CT which significantly increases the scanning resolution: current86
generation XRCT devices are able to achieve maximum resolutions of roughly 0.05 µm and87
resolutions of 10 µm for objects that are a few millimetres across (Rigby et al., 2011).88
XRCT uses X-ray attenuation (the loss of energy as a ray passes through a material) to89
reconstruct images of the interior of an object (Keller, 1998). Radiographic projections (i.e. 2-90
D images acquired using X-rays) of the object are taken from many angles by rotating the object91
relative to a stationary X-ray source and detector. The X-ray attenuation is then calculated92
at specific points and the 2-D image or slice constructed. 3-D reconstruction of the object can93
be achieved by converting slice pixels into voxels and then “stacking” several slices on top of94
each other. The reader is referred to Kruth et al. (2011) for a detailed description of the XRCT95
process.96
XRCT was first used with geotechnical materials to observe 2-D stress fields in sands (e.g.97
Roscoe (1970)). 3-D studies on strain localisation in sands have been conducted since the late98
1980s, achieved using XRCT images and subsequent image analysis techniques (for example99
particle image velocimetry, as discussed in Beckett and Augarde (2011)) to track deformations100
3
during loading (Colliat-Dangus et al., 1988; Van Geet et al., 2000; Desrues and Viggiani, 2004;101
Hall et al., 2010). XRCT has recently been used to characterise the hydration of clay pellets102
and powders (Van Geet et al., 2005; Gens et al., 2011).103
Several imperfections which arise during XRCT scanning were identified by Van Geet et al.104
(2000):105
• Ring artifacts are caused by inhomogeneities in the projector. They can be detected and106
accounted for by randomly moving the object relative to the detector.107
• Star artifacts are caused by very dense inclusions within the object, radiating incident108
rays back out of the specimen. Star artifacts can be reduced by placing different filters in109
front of the projector, depending on the material causing the error.110
• Shadowing occurs due to the absorbing of low-energy waves at the edges of objects. Again,111
the use of filters can prevent shadowing.112
3 Fractal analysis113
Fractal analysis can be used to describe the shapes and arrangements of irregular objects (Man-114
delbrot, 1967). As opposed to Euclidean geometry, where objects are classified as one, two or115
three dimensional, the dimensionality of an irregular line can take any value between one and116
two, with more irregular lines having larger fractal dimensions. As fractals quantify irregular117
structures, they are a suitable tool for quantifying the internal structure of RE and changes118
which occur to it. Fractals have been used to investigate several properties of geotechnical119
materials, for example in modelling particles and pore networks (Perfect et al., 1992; Lipiec120
et al., 1998; Perrier et al., 1999; Perrier and Bird, 2002; Atzeni et al., 2008), permeability (The-121
vanayagam and Nasarajah, 1998; Xu and Sun, 2002; Xu, 2004a; Cihan et al., 2009; Jobmann and122
Billaux, 2010) and retention curve properties (Bird et al., 1996; Gimnez et al., 1997; Kravchenko123
and Zhang, 1998; Huang et al., 2006; Russell, 2010). Fractals have also been used to describe124
soil cohesive properties (Bonala and Reddi, 1999) and unsaturated shear strength (Xu, 2004b).125
The fractal dimension of a line can be found by dividing it into elements of length λ (known126
as the dividing method) and finding the number of such elements, P (λ), that are required to127
cover the line. P (λ) and λ are related by128
P (λ) = kλ−Df (1)
4
where k is a constant and Df is the fractal dimension (Ersahin et al., 2006). Df is found by129
plotting λ on the abscissa axis against P (λ) on the ordinate axis using a logarithmic scale. The130
gradient of the resulting linear fit (necessary for the line to be considered fractal), m, is related131
to Df via Df = −m. The fractal dimension of a straight line or circular arc will always be 1132
using this method (Vallejo, 1996).133
The box-counting method overlays an object with parallel vertical and horizontal lines, sep-134
arated by a distance λ (Hirata, 1989). Each “box” formed by the intersecting lines forms one135
element and P (λ) is found by counting the number of boxes which contain part of the object,136
as shown in Figure 1. Successively fine grids are used to determine Df from a plot of λ against137
P (λ) as before. The accuracy of the estimate of Df can be improved by choosing an appropri-138
ate range of values for λ and by using multiple grid locations, using the average of the number139
of boxes filled for each location to find Df (Foroutan-pour et al., 1999). The minimum cover140
value for Df is the fractal dimension calculated using the grid which requires the least number141
of elements to cover the object (Tolle et al., 2003). Therefore, if only one grid location is used,142
the average and minimum cover Df values are the same. Like the dividing method, the fractal143
dimension for a straight line or circular arc found using the box-counting method will always144
be 1.145
The fractal dimension gives a measure of the size and dispersivity of objects. However, Df146
cannot indicate the amount of objects present in the pattern: therefore, an additional parameter147
is required to describe fractal patterns. The lacunarity, L, represents the interconnectivity (or148
heterogeneity or gappiness) of the pattern, and can be considered to be the P (λ) intercept of149
the linear fit to the plot of λ against P (λ). A larger value of L therefore indicates that there150
are more gaps between the objects so that a greater number of boxes is required to cover the151
pattern (Zeng et al., 1996). Both the lacunarity and the fractal dimension are therefore required152
to describe a fractal pattern (Pachepsky et al., 2000; Blair et al., 2007).153
4 Experimental procedure154
4.1 Soil mix selection and preparation155
It is recommended that specimens be of the order of 1000 times larger than the desired resolu-156
tion for XRCT scanning (Ketcham and Carlson, 2001). Specimens of 20 mm height and 15 mm157
diameter were selected in order to use a high scanning resolution (1 pixel to 11 µm) whilst being158
5
large enough to be representative of material used in construction. Although a resolution of159
11 µm is insufficient to investigate the smallest pores within the material (on the nanometre160
scale), it is suggestibly sufficient to determine structural changes due to crack formation. There-161
fore, that this study focuses on changes which occur to the material macrostructure (with pores162
identified necessarily belonging to the macropore class). As cracking in quasi-brittle materials,163
for example RE, is a macroscopic phenomenon, this resolution should be sufficient to determine164
the effects of loading and compaction on material macrostructure (Dehandschutter et al., 2005).165
A soil mix comprising 70% sand and 30% silty-clay by mass was selected for testing. Al-166
though traditional RE soil mixes should contain a gravel fraction (Houben and Guillaud, 1996),167
the restriction on specimen size precludes the incorporation of large particles; the material in-168
vestigated here can therefore be considered to be representative of the intra-aggregate (or ”soil169
matrix”) material, rather than the entire RE mix.170
The mix was prepared by combining dry constituent materials in the appropriate propor-171
tions. Dried silty clay (“Birtley clay”, LL 58.8%, PL 25.7%, PI 33.1%) was prepared by drying172
lumps of the material at 105◦C for 48 hours (BS1377:1990), prior to being pulverised and passed173
through a 2.36 mm sieve. This sieve size was selected as particle aggregates were small enough174
to mix uniformly with the remaining mix fractions on the creation of the soil mix whilst pro-175
ducing a sufficient quantity of material at an acceptable rate. Dried sand was also sieved to176
pass 2.36 mm in order to remove gravel-sized particles (2–20 mm, BS1377:1990). The light177
Proctor test was used to determine the mix optimum water content (OWC); this was chosen as178
it is deemed representative of the compaction applied to full-scale RE structures (Beckett and179
Augarde, 2010). Material maximum bulk and dry densities and OWC are given in Table 1.180
4.2 Specimen preparation and testing181
Specimens were manufactured using the apparatus shown in Figure 2. The mould shown in182
Figure 2 is open at both ends to facilitate extraction; during manufacture, the mould was183
secured to a removable steel base plate (not shown in Figure 2). Twelve single– and three184
double-layer specimens were prepared for XRCT scanning. Consecutive testing on a single185
specimen, for example as performed by Hall et al. (2010) and Ando et al. (2012), was not186
possible due to limitations on scanner availability. This is acceptable as cracking within the187
material do not close on the removal of an applied load (Atkinson, 1987).188
Wet material was prepared in a sealable polypropylene container in three layers of 2.5 kg,189
6
wetted to OWC using distilled water. The container was treated with an anti-static cleaner prior190
to being filled with the dry soil mix to prevent dry clay particles from adhering to the container191
surfaces and sealed for 48 hours to allow the water to equilibrate. Single-layer specimens were192
manufactured using the single-layer press (Figure 2) and static compaction, with the amount193
of material required determined through the required density and known specimen volume.194
Double-layer specimens were manufactured by allocating half of the required material to each195
layer and compacting the first layer with the double-layer press and the second with the single-196
layer press.197
Specimens were placed on wire racks to dry naturally to a constant mass at a tempera-198
ture of 20◦C ±2◦C and a relative humidity of 40% ±5%. The porosities of the single– and199
double-layer specimens were 0.290 (standard deviation of 0.006) and 0.263 (standard deviation200
of 0.001) respectively, determined from the specimen mass, dry density and volume on reach-201
ing a constant water content. Two single-layer and two double-layer specimens (numbers 2s,202
10s, 13d and 15d, where “s” and “d” denote single– and double-layer specimens respectively)203
were selected for XRCT testing as they had the most similar dry masses and so, assumedly,204
similar material structures. The remaining specimens were used to determine the unconfined205
compressive strength (UCS) of the material by being crushed between low-friction plattens at206
a set displacement rate of 0.5 mm/min. The average specimen compressive failure load was207
287.4 N ±4.2 %, corresponding to an average UCS of 1.63 MPa, calculated as the applied load208
over the specimen cross-sectional area. Correction factors for the compressive strength based209
on the specimen aspect ratio have been suggested by a number of authors (e.g. Krefeld (1938)210
and Heathcote and Jankulovski (1992)), however it is suggested in Morel et al. (2007) that they211
are not suitable for use with earthen materials; UCSs have therefore been left unmodified.212
Formwork used in RE construction is usually limited to 1.5 m high, in order to facilitate its213
movement around the site (King, 1996). If it is assumed that the wet density RE is roughly214
2.1 Mg/m3 (following results from Beckett and Augarde (2011)) and a layer, once compacted,215
is roughly 100 mm deep, then it can be shown that the bottommost layer is subjected to a216
load of roughly 29 kN/m2 due to the material self weight, or roughly 2% of the average UCS217
reported above. As this is very small, it can be assumed that no significant cracking occurs218
due to material self weight. Therefore, the macrostructures of small specimens, as opposed to219
samples taken from a larger wall, should be adequate to determine changes which occur to the220
macrostructure of material present in full-scale RE walls on loading. Loads corresponding to221
7
85% and 25% of the average failure load were then applied to specimens 2s and 10s respectively;222
these values were selected to provide a sufficient difference between their loaded macrostructures223
whilst enabling specimens to be safely transported. Specimens 13d and 15d were not loaded. A224
summary of these specimen preparation stages is given in Table 2225
5 Results and discussion226
5.1 Image processing227
XRCT images were obtained using a Phoenix Nanotom 180NF XRCT system (GE Sensing and228
Inspection Technologies, GmbH, Wunsdorf, Germany) at a resolution of 1 pixel = 11 µm. Each229
set of projection images was reconstructed using the back projection algorithm in the “datos|x230
rec” software. “ImageJ v.1.43u” software (Rasband, 2002) was then used to process the XRCT231
images.232
Examples of the 3-D detected pore networks, determined using the IsoData thresholding233
values given in Table 2 and produced using the AvizoFire software (Visual Sciences Group,234
Burlington, USA), are shown in Figures 3 and 4, which give an impression as to the highly235
complex nature of the macropore spaces within these specimens. Due to this complexity, 2-D236
image slices, rather than 3-D interpolations, are used here to determine changes that occur to237
the macrostructure on loading, using the procedures discussed above. An example of such a238
slice is shown in Figure 5(a), along with the specimen axes as defined by the XRCT scanning239
software, as well as the image prior to and post-processing using ImageJ; pores are shown as240
black pixels in the processed images, whilst solid material is shown as white. Example outlines241
of particles seen in Figure 5(b) are shown in Figure 5(c) for reference. Slices in the XY plane are242
used as this is the plane perpendicular to compaction, and it is expected that macrostructural243
changes will occur with depth, but not in the radial direction. The image processing procedure244
is given in Figure 6.245
Images are divided into pore spaces and solid material using a threshold intensity value; the246
lower the value, the lower the pixel intensity (i.e. the darker the shade) that corresponds to247
pores. In order to prevent shadowing, the upper and lower 100 images, corresponding to the248
upper and lower 1.1 mm, were removed from the image sequence. The removal of these images249
also reduces the chance of any damaged material, likely to be found at the ends of the specimen,250
from influencing the analysis. Failure to remove these images results in a lower thresholding251
8
value being selected, so that fewer pores are identified. Roughly 1600 images in the XY plane252
were therefore subsequently available for analysis per specimen, at 11 µm intervals. An area253
of 938 × 856 pixels was cropped from the centre of XY-orientated images to avoid shadowing254
at the specimen edges and to provide a constant cross sectional area for the analysis. Due to255
the large number of images, only one in every ten images (i.e. one image every 110 µm) was256
processed.257
For thresholding, original image brightness values were modified whereby the lowest intensity258
present in the image was set to zero and the highest to 255 (the minimum and maximum259
possible intensity values respectively), in order to ensure the greatest contrast between pixels.260
A “median” filter was then applied in order to reduce the noise in the image by removing261
outlying intensity regions. This method replaces a pixel’s intensity with the median intensity262
of a region of a given size centred on that pixel (a region of a 2 pixel radius was used in this263
case). The brightness values were then corrected again, using the process described above, to264
account for the filtering process. The ImageJ “IsoData” thresholding algorithm was applied to265
the modified images to determine a suitable thresholding value.266
The IsoData process divides the image into pores and solids using an initial threshold value.267
The average intensities of those pixels whose intensities are above and below that value are then268
calculated and the average of those is used as a second threshold value. The process is repeated269
until the calculated threshold is larger than the average intensity for the entire image (Ridler and270
Calvard, 1978). Double-layer specimens, although their layers were scanned separately, were271
analysed as one image sequence in order to directly compare one layer to another. Final IsoData272
threshold values are given in Table 2, with the same thresholding value being applied to every273
image in the image sequence. Values differ between specimens due to the manual brightness274
correction procedure and subtle differences in their structure (e.g. particle mineralogy) which275
influence that correction.276
5.2 Effect of loading on material macrostructure277
Results found for specimens 2s and 10s were used to determine the effect of loading on material278
macrostructure. Preliminary testing indicated that the use of five grid positions with a minimum279
box size of 1 pixel gave the most repeatable results for macrostructure fractal properties using the280
minimum cover box counting method. The Pore Area Fraction (PAF, calculated by dividing the281
detected pore area by the total image area), fractal dimension, Df , and lacunarity, L, calculated282
9
for each specimen using the processed images, for a range of tested image thresholding values,283
are shown in Figures 7 to 9 respectively. Depths have been calculated assuming a 110 µm284
separation between images and have been corrected to account for the 1.1 mm of material285
removed from the top and bottom of the image sequence. To prevent the loss of loose particles286
from the base of the specimens and to ensure a good seating within the scanner, specimens287
were placed in the scanner with the topmost compacted surface (which is the smoothest) facing288
downwards. Therefore, a depth of 0 in Figures 7 to 9 corresponds to the base of the specimen.289
A minimum pore size of 3 × 3 pixels (i.e. ≥1089µm2) was used to calculate the PAF in order to290
prevent anomalous features (for example small pits on particle surfaces or random image noise)291
from influencing the results, however a minimum box size of 1 pixel was used to determine Df292
and L (i.e. features ≥121µm2 are included in the fractal analysis).293
Figures 7 and 8 show a good level of continuity between consecutive values of PAF and Df294
found using IsoData thresholding intensity, suggesting that a resolution of 11 µm is suitable for295
investigating the macrostructures of these materials. These figures also show a sudden increase296
in PAF and Df , for both specimens, at depths near 0 and 20 mm. However, Figures 10(a) and297
(b), taken at depths of 16500 µm and 17379 µm respectively for specimen 2s (i.e. correspond-298
ing to regions just below and at the tip of these sudden increases), show very little apparent299
difference between their macrostructures. These sudden increases are therefore suggested to be300
indicative of shadowing, whereby slightly darker image at the specimen edges give unrealistically301
large detected pore areas as compared to images taken from the bulk of the specimen. Ignoring302
these shadowed regions, the maximum measured PAFs detected for specimens 2s and 10s at the303
IsoData threshold value are roughly 0.16 and 0.14 respectively, which are significantly smaller304
than their average bulk porosities of 0.29. This is again indicative of how XRCT can only be305
used to gain information about the material macrostructure using this resolution.306
IsoData values shown in Figure 7 clearly indicate that PAF, and so by extension macrostruc-307
tural density and material strength (Hall and Djerbib, 2004), increases with increasing depth308
(i.e. towards the top of the specimen). A slight reduction in Df and an increase in L is also309
seen in Figures 8 and 9 with increasing depth, suggesting that detected macropores become310
elongated and more isolated, supporting the observation that the material is more heavily com-311
pacted towards the top of the specimen. This is unexpected, as the small depth of the layer312
would suggestibly be sufficient to ensure uniform compaction throughout the material. Instead,313
results suggest that friction with the sides of the mould results in this decreasing density to-314
10
wards the specimen base (Morel et al., 2007). Whether such a result would occur in full-scale315
RE formwork (of the order of 300 mm wide) is unclear, due to the different friction conditions316
and significantly larger volumes involved, however it is clear that care should be taken to ensure317
that specimen compaction results in a similar material to that found on-site.318
The nature of the macrostructures of specimens 2s and 10s can be investigated in more319
detail by varying the threshold values (low thresholding values detect the darkest pixels and so320
the centres of the largest pores, whilst smaller pores are detected as the value is increased), as321
the arrangement of large and small detected macropores can be determined (Taud et al., 2005).322
An example of the use of different threshold intensities to investigate an image slice is shown in323
Figure 11. Detected PAF, Df and L values using different threshold intensities are also shown324
in Figures 7 to 9 (selected intensities are shown in parentheses in the figure legends). Threshold325
values below 50 were not examined as no pores could be detected in some images. Similarly,326
values larger than the IsoData threshold values (Table ??) were not investigated as some images327
became completely populated by pores.328
Figure 7 shows that higher values of PAF and Df , and lower values of L, are largely observed329
for specimen 2s than for specimen 10s at all investigated threshold intensity values and sample330
depths. An increasing Df and reducing L with increasing threshold value is indicative of331
smaller pores residing between, and not in isolation of, larger pores, so that smaller features332
(macropores and cracks) propagate from larger features. The similar trends for PAF, Df and L333
for both specimens suggest that their initial macrostructures were similar, so that the differences334
between them are due to the application of differing loads only. Therefore, results shown in335
Figures 7 to 9 for different threshold values suggest that the application of a load to the initial336
macrostructure affects the entire pore size range investigated, and that it results in the formation337
and propagation of cracks from the larger to the smaller pores. This has important consequences338
for considerations of hydraulic conductivity, for example, as increasing pore interconnectivity339
will result in increased conductivity.340
As results discussed above suggest that the initial macrostructures of specimens 2s and 10s341
were indeed similar, the extent by which pores of different sizes are affected by loading can be342
determined by examining the relative changes in PAF, Df and L between specimens at given343
depths. Figures 12 to 14 show the percentage differences between PAF, Df and L for the two344
11
specimens respectively, calculated using345
value 2s − value 10s
value 2s× 100. (2)
Note that Eqn ?? removes the effect of shadowing. Although PAF reduces with increasing346
sample depth (Figure 7), Figure 12 shows that, with some oscillations (which are discussed347
below), the relative change in PAF between the two samples is relatively constant for a given348
depth at around +30% for all threshold intensities, suggesting that pores of all sizes were equally349
affected by the application of a load. By extension, this suggests that material compacted to350
a lower density will experience greater cracking than the same material compacted to a higher351
density, due to the greater number of pre-existing flaws. This is consistent with the theory352
of cracking of quasi-brittle materials, as it is expected that a greater number of cracks can353
propagate from larger initial flaws (Atkinson, 1987). Figures 13 and 14 show that IsoData values354
of Df and L remained relatively constant at +5% and −50% respectively, again indicating that355
flaws (macropores and cracks) become larger and more interconnected on the application of a356
load and again consistent with the above theory. A greater variation in relative PAF, Df and357
L values is seen for lower threshold values; this is due to the more isolated arrangements of the358
identified features at these intensity values, again as shown in Figure 11.359
Figures 12 and 13 show several oscillations in obtained results between depths of 0 and360
roughly 12 mm, due to similar oscillations seen in Figures 7 and 8. As the oscillations for361
each specimen are not in phase with each other with depth, which would suggest a material362
property, this instead suggests that errors were encountered whilst moving the specimen relative363
to the X-Ray source and receptor, perhaps due to an incomplete isolation of the specimen from364
external light. However, it is assumed here that these errors do not detract from the usefulness365
of the data to determine structural changes on the application of a load.366
Figures 12 and 13 show a large shift to negative relative PAF and Df values between the367
depths of 12 and 17 mm, due to an increase in PAF and Df of specimen 10s between these368
depths, apparently contradicting results discussed above. An image slice taken from this region369
of sample 10s (at a depth of 13090 µm) is shown in Figure 11, showing that a large particle370
(roughly 10 mm) is present at this depth which was not removed during sieving. The effect of371
this particle is to reduce the compactivity of that material adjacent to it, as shown in Figures 7372
and 8 through the increase in PAF and Df in this region. A considerable increase in L is373
12
also seen in this depth range, due to the large ‘gap’ which the particle represents. It should374
be noted, however, that although the relative differences between specimens 2s and 10s in this375
region are large, the absolute differences, as shown in Figures 12 to 14, are quite small as376
compared to the largest values determined (for example, the highest PAF found for specimen377
10s, excluding shadowed areas, is roughly 15%, whereas PAF in the region containing the large378
ranges between only 5 and 10%). Results found for specimen 10s are therefore suggestibly still379
valid for determining the effects of loading on material macrostructure, although relative values380
are misrepresentative of these effects in the region of the large particle.381
5.3 Effect of compaction on the material macrostructure382
Specimens 13d and 15d were used to investigate changes which occur to material macrostructure383
on the compaction of additional layers. Values of PAF, Df and L determined for each specimen384
using different threshold values are shown in Figures 15 to 17 respectively. Again, a depth of385
20 mm corresponds to the top of the specimen due to its orientation in the XRCT scanner.386
Opposite to Figures 7 and 8, Figures 15 and 16 show reduced PAF and Df values at the ends387
and in the centres of the specimens, whilst Figure 17 shows higher L values at the specimen388
ends, suggesting that images taken from these regions were in fact darker than those taken from389
the bulk of the material (i.e. a reversed shadowing effect, as a darker image returns a greater390
number of detected pores). PAF and Df values in Figures 15 and 16 are also significantly391
higher than those shown in Figures 7 and 8, suggesting that the majority of the space in the392
image slices has been incorrectly classified as pore space (as a PAF value of 1, or Df value393
of 2, represents a rectangular area entirely comprising pore space). These results suggest that394
the IsoData thresholding method is inappropriate for analysing these specimens. Examples of395
the use of different thresholding values on detected pore area are shown in Figure 18; although396
The IsoData threshold value results in a realistic approximation to the pore network shown in397
Figure 18(a) (taken from the top of the specimen, depth 1100 µm) in Figure 18(c), it severely398
overestimates the pore area of Figure 18(d) (taken from the bulk material, depth 17600 µm) in399
Figure 18(f), and hence produces the drops in PAF and Df values seen at the specimen ends400
and centre in Figures 15 and 16.401
Due to the unrealistic PAF and Df and L values obtained using the IsoData threshold402
value, images were re-analysed using a lower threshold value of 60 (chosen following preliminary403
work). These results are also shown in Figures 15 to 17. Figure 18 shows that the use of404
13
the lower threshold intensity of 60 produces a far better approximation to the pore network in405
Figure 18(b), but fails to capture most of the pores in Figure 18(e). Given that results discussed406
in the previous section show that results for the uppermost and lowermost material should be407
discounted due to shadowing (or the apparent ’reversed shadowing’ in this case), the use of408
a threshold value of 60 appears to be the more appropriate for investigating changes to the409
macrostructure on layer compaction.410
Figures 15 to 17 show relatively good agreement between PAF, Df and L values determined411
using the threshold value of 60 for both specimens, suggesting that their macrostructures are412
quite similar and so that results for both can be used to investigate the effects of layer compaction413
on material macrostructure. The effect of shadowing detected at this threshold value is also414
consistent with that found in Figures 7 to 9 (i.e. an increase in the number of detected pores415
at the edges), suggesting that the ‘reversed’ shadowing in Figures ?? to ?? was indeed due416
to the use of too high a threshold value. As in Figures 12 and 13, oscillations are also seen in417
Figures 15 and 16 for changing PAF and Df with depth respectively. However, these oscillations418
are in phase with each other for specimens 13d and 15d, again suggesting that they are due to419
an error encountered whilst moving specimens within the scanner during the analysis and not420
due to a change in the macrostructure. This further suggests that the change in the phase of421
these oscillations between results for specimens 2s and 10s in Figures 15 and 16 is the result of422
a change in the macrostructures of those specimens due to loading.423
A considerable change in seen in the values found for PAF, Df and L for both specimens424
at a depth of roughly 9 mm in Figures 15 to 17, whereby material in the upper layer has425
considerably higher PAF and Df values and lower L values than that in the bottom layer, so426
that the material in the upper layer is characterised by larger macropores with a higher degree427
of interconnectivity. This is supported by Figure 19, which shows a central section, in the428
XZ plane, through sample 13d and which clearly shows that the upper compacted layer has a429
significantly higher porosity than the lower layer (layers are indicated by boxes in Figure 19). Of430
interest, however, is the placement of the layer interface at a depth of 9 mm from the uppermost431
surface, as opposed to the intended 10 mm. If this were due to the additional compaction of the432
underlying layer, then it would be expected, following results seen in Figure 7, that a density433
gradient would be observed throughout the layer, with density reducing with increasing layer434
depth. The identical depths of the layer interfaces in both specimens would also not be expected435
if the change in material structure were due to the additional compaction of the underlying layer.436
14
These observations therefore suggest that the significant difference in the macrostructures of437
the two layers is due to an error in the length of the double-layer press, in that it was infact438
11 mm, as opposed to the intended 10 mm, which resulted in an over-compaction of the specimen439
material in the bottom layer, due to the use of a set mass of material, and a subsequent under440
compaction of the material in the upper layer. This therefore suggests that the macrostructure of441
the material in the underlying layer was not, in fact, affected by the compaction of the material442
above it, supporting results presented in (Beckett and Augarde, 2011), although further testing443
is required to confirm this due to the under-compacted nature of the upper layer. Therefore, the444
result found in the previous section that macrostructural density reduces, with an associated445
increase in macropore size and interconnectivity, towards the base of a compacted layer can be446
applied to every layer within an RE structure.447
Morel et al. (2007) observed that compressive strength correction factors for different aspect448
ratios determined for masonry materials are not suitable for soils, as the material is not homo-449
geneous throughout the layer. However, they assumed that the material comprising specimens450
with aspect ratios below 1.5 can be considered homogeneous, so that existing correlation factors451
might be acceptable. Furthermore, the specification of an aspect ratio does not place limita-452
tions on any one dimension of a specimen, depending on the values of the other dimensions.453
Results presented here show that the material comprising the tested single-layer samples (as-454
pect ratio4
3) is highly heterogeneous, with largely homogeneous material only observed in the455
even shallower double-layer specimen layers (aspect ratio2
3, although the compaction applied456
to these layers is suspect). Therefore, it is suggested that layer thickness, and by extension the457
relation between that and maximum particle size, be considered when determining the degree458
of homogeneity of RE materials as opposed to specimen aspect ratio.459
6 Conclusion and recommendations460
The aims of this investigation were to use XRCT scanning to observe the changes that occurred461
to the macrostructure of RE on loading and to determine how the the macrostructure changed462
throughout a layer as a result of compaction.463
Results presented for all specimens suggest that an XRCT scanning resolution of 11 µm464
is sufficient to investigate macrostructural properties of RE materials. Results for single-layer465
specimens have shown that the application of a load to an RE material results in a significant466
15
increase in the size and interconnectivity of macropores due to cracking, with the degree of467
cracking increasing for larger loads. It was also shown that the effect of cracking was greater468
for larger macropores than for smaller, due to the greater number of cracks that can propagate469
from the larger. Therefore, poorly-compacted material will suffer more cracking than well-470
compacted material for the same applied load, indicating the need for proper compaction during471
construction.472
Results for single-layer specimens show that macrostructural density continuously reduces473
towards the base of a compacted layer. It is therefore likely that material found at the bottom474
of a compacted layer is weaker than that at the top, with a higher hydraulic conductivity due475
to increased pore size and interconnectivity.476
Results found for double-layer specimens have shown that the macrostructure of a compacted477
layer is not affected by the compaction of additional material above it. Therefore, properties478
of RE specimens comprising few compacted layers can be considered to be representative of479
material contained within full-scale RE structures, although layer thickness should be similar480
to that found in the full-scale structure due to the variations in material density with layer481
depth identified above.482
Results for single-layer specimens show that the macrostructure of material comprising a483
compacted layer is highly heterogeneous, with largely homogeneous material only found in the484
layers of double-layer specimens. Care should therefore be taken when specifying minimum485
specimen aspect ratios to achieve material homogeneity. Instead, limitations should be placed486
on specimen layer thickness and maximum particle size.487
7 Acknowledgements488
The first author is supported by a studentship awarded by the School of Engineering and489
Computing Sciences, Durham University.490
16
References491
Ando, E., Hall, S., Viggiani, G., Desrues, J., Besuelle, P., 2012. Grain-scale experimental in-492
vestigation of localised deformation in sand: a discrete particle tracking approach. Acta493
Geotechnica 7 (1), 1–13.494
Atkinson, B. K. (Ed.), 1987. Fracture mechanics of rock. Academic Press Inc., London (UK).495
Atzeni, C., Pia, G., Sanna, U., Spanu, N., 2008. A fractal model of the porous microstructure496
of earth-based materials. Construction and Building Materials 22 (8), 1607–1613.497
Baltodano-Goulding, R., September 2010. Deformation behavior in unsaturated soils. In:498
Alonso, E., Gens, A. (Eds.), Unsaturated Soils. Fifth International Conference on Unsat-499
urated Soils, CRC Press, pp. 181–185.500
Beckett, C. T. S., 2011. The role of material structure in compacted earthen building materials:501
Implications for design and construction. PhD Thesis, Durham University.502
Beckett, C. T. S., Augarde, C. E., September 2010. Development of microstructure in com-503
pacted earthen building materials. In: Alonso, E., Gens, A. (Eds.), Unsaturated Soils. Fifth504
International Conference on Unsaturated Soils, CRC Press, pp. 139–144.505
Beckett, C. T. S., Augarde, C. E., 2011. A novel image-capturing technique for the experimental506
study of soil deformations under compaction. Geotechnical Testing Journal 34 (6), 571–578.507
Betts, M. C., Miller, T. A. H., 1937. Rammed earth walls for buildings. Farmers Bulletin No.508
1500. U.S. Department of Agriculture, Washington D.C. (U.S. Govt. Printing Office).509
Bird, N., Bartoli, F., Dexter, A., 1996. Water retention models for fractal soil structures. Euro-510
pean Journal of Soil Science 47, 1–6.511
Blair, J. M., Falconer, R. E., Milne, A. C., Young, I. M., Crawford, J. W., 2007. Modeling512
three-dimensional microstructure in heterogeneous media. Soil Science Society of America513
Journal 71 (6), 1807–1812.514
Bonala, M. V. S., Reddi, L. N., 1999. Fractal representation of soil cohesion. Journal of Geotech-515
nical and Geoenvironmental Engineering 125 (10), 901–904.516
Braunack, M. V., Hewitt, J. S., Dexter, A. R., 1979. Brittle fracture of soil aggregates and517
compaction of aggregate beds. Journal of Soil Science 30, 653–667.518
17
BSI, 1990. BS 1377:1990. Methods of test for soils for civil engineering purposes.519
Cihan, A., Tyner, J. S., Perfect, E., 2009. Predicting relative permeability from water retention:520
A direct approach based on fractal geometry. Water Resources Research 45, 8.521
Colliat-Dangus, J. L., Desrues, J., Foray, P., 1988. Triaxial testing of granular soil under elevated522
cell pressure. In: Donaghe, R. T., Chaney, R. C., Silver, M. L. (Eds.), Proceedings of a523
conference on advanced triaxial testing for soil and rocks. American Society for Testing and524
Materials, Philadelphia (USA), pp. 290–310.525
Collins, K., McGown, A., 1974. Form and function of microfabric features in a variety of natural526
soils. Geotechnique 24 (2), 223–254.527
Dehandschutter, B., Vandycke, S., Sintubin, M., Vandenberghe, N., Gaviglio, P., Sizun, J. P.,528
Wouters, L., 2004. Microfabric of fractured boom clay at depth: a case study of brittle-ductile529
transitional clay behaviour. Applied Clay Science 26 (1-4), 389–401.530
Dehandschutter, B., Vandycke, S., Sintubin, M., Vandenberghe, N., Wouters, L., 2005. Brittle531
fractures and ductile shear bands in argillaceous sediments: inferences from oligocene boom532
clay (belgium). Journal of Structural Geology 27 (6), 1095–1112.533
Desrues, J., Viggiani, G., 2004. Strain localization in sand: an overview of the experimental534
results obtained in grenoble using stereophotogrammetry. International Journal for Numerical535
and Analytical Methods in Geomechanics 28, 279–321.536
Diamond, S., 1971. Microstructure and pore structure of impact-compacted clays. Clays and537
Clay Minerals 19, 239–249.538
Ersahin, S., Gunal, H., Kutlu, T., Yetgin, B., Coban, S., 2006. Estimating specific surface area539
and cation exchange capacity in soils using fractal dimension of particle-size distribution.540
Geoderma 136 (3-4), 588–597.541
Foroutan-pour, K., Dutilleul, P., Smith, D. L., 1999. Advances in the implementation of the542
box-counting method of fractal dimension estimation. Applied Mathematics and Computation543
105 (2–3), 195–210.544
Gens, A., Vallejan, B., Sanchez, M., Imbert, C., Villar, M. V., Van Geet, M., 2011. Hydrome-545
chanical behaviour of a heterogeneous compacted soil: experimental observations and mod-546
elling. Geotechnique 61 (5), 367–386.547
18
Gimnez, D., Perfect, E., Rawls, W. J., Pachepsky, Y., 1997. Fractal models for predicting soil548
hydraulic properties: a review. Engineering Geology 48 (3-4), 161–183.549
Hall, M., Djerbib, Y., 2004. Rammed earth sample production: context, recommendations and550
consistency. Construction and Building Materials 18 (4), 281–286.551
Hall, S. A., Bornert, M., Desrues, J., Pannier, Y., Lenoir, N., Viggiani, G., Besuelle, P., 2010.552
Discrete and continuum analysis of localised deformation in sand using x-ray µct and volu-553
metric digital image correlation. Geotechnique 60 (5), 315–322.554
Heathcote, K., Jankulovski, E., 1992. Aspect ratio correction factors for soilcrete blocks. Aus-555
tralian civil engineering transactions CE34 (4: Australia: Institution of Engineers), 309–312.556
Hirata, T., 1989. Fractal dimension of fault systems in Japan: Fractal structure in rock fracture557
geometry at various scales. Pure and Applied Geophysics 131, 157–170.558
Houben, H., Guillaud, H., 1996. Earth construction - a comprehensive guide., Second Edition.559
Intermediate Technology Publications, London (UK).560
Hoyos, L. R., Perez-Ruiz, D., Puppala, A. J., 2010. Constitutive modeling of unsaturated soil561
behavior using a refined suction-controlled true triaxial cell: Preliminary observations. In:562
Alonso, E., Gens, A. (Eds.), Unsaturated Soils. Fifth International Conference on Unsaturated563
Soils, CRC Press.564
Huang, G.-H., Zhang, R.-D., Huang, Q.-Z., 2006. Modeling soil water retention curve with a565
fractal method. Pedosphere 16, 137–146.566
Jaquin, P. A., Augarde, C. E., Gallipoli, D., Toll, D. G., 2009. The strength of unstabilised567
rammed earth materials. Geotechnique 59 (5), 487–490.568
Jobmann, M., Billaux, D., 2010. Fractal model for permeability calculation from porosity and569
pore radius information and application to excavation damaged zones surrounding waste570
emplacement boreholes in opalinus clay. International Journal of Rock Mechanics and Mining571
Sciences 47 (4), 583–589.572
Keller, A., 1998. High resolution, non-destructive measurement and characterization of fracture573
apertures. International Journal of Rock Mechanics and Mining Sciences 35 (8), 1037–1050.574
19
Ketcham, R. A., Carlson, W. D., 2001. Acquisition, optimization and interpretation of x-ray575
computed tomographic imagery: Applications to the geosciences. Computers and Geosciences576
27, 381–400.577
King, B., 1996. Buildings of Earth and Straw: Structural Design for Rammed Earth and Straw-578
Bale Architecture. Ecological Design Press, Sausalito, California (USA).579
Kravchenko, A., Zhang, R., 1998. Estimating the soil water retention from particle-size distri-580
butions: a fractal approach. Soil Science 163 (3), 171–179.581
Krefeld, 1938. Effect of shape of specimen on the apparent compressive strength of brick ma-582
sonry. In: Proceedings of the American Society of Materials, Philadelphia, USA. pp. 363–369.583
Kruth, J., Bartscher, M., Carmignato, S., Schmitt, R., Chiffre, L. D., Weckenmann, A., 2011.584
Computed tomography for dimensional metrology. CIRP Annals - Manufacturing Technology585
60 (2), 821 – 842.586
Lipiec, J., Hatano, R., Slowinska-Jurkiewicz, A., 1998. The fractal dimension of pore distribution587
patterns in variously-compacted soil. Soil and Tillage Research 47 (1-2), 61–66.588
Mandelbrot, B. B., 1967. How long is the coast of Great Britain? Statistical self-similarity and589
the fractional dimension. Science 156, 636–638.590
Monroy, R., Zdravkovic, L., Ridley, A., 2010. Evolution of microstructure in compacted London591
clay during wetting and loading. Geotechnique 60, 105–199.592
Morel, J.-C., Pkla, A., Walker, P., 2007. Compressive strength testing of compressed earth593
blocks. Construction and Building Materials 21 (2), 303–309.594
Pachepsky, Y. A., Gimnez, D., Crawford, J. W., Rawls, W. J., 2000. Conventional and fractal595
geometry in soil science. In: Y. Pachepsky, J. C., Rawls, W. (Eds.), Fractals in Soil Science.596
Vol. Volume 27. Elsevier, pp. 7–18.597
Perfect, E., Rasiah, V., Kay, B. D., 1992. Fractal dimensions of soil aggregate-size distributions598
calculated by number and mass. Journal of the Soil Science Society of America 56, 1407–1409.599
Perrier, E. M. A., Bird, N. R. A., 2002. Modelling soil fragmentation: the pore solid fractal600
approach. Soil & Tillage Research 64, 91–99.601
20
Perrier, E. M. A., Bird, N. R. A., Rieu, M., 1999. Generalizing the fractal model of soil structure:602
the poresolid fractal approach. Geoderma 88, 137–164.603
Proctor, R. R., 1933. Fundamental principles of soil compaction. Engineering News Record604
111 (9), 245–248.605
Rasband, W., 2002. NIH Image J.606
URL http://rsbweb.nih.gov/ij/index.html (verified 02.08.2011)607
Ridler, T. W., Calvard, S., 1978. Picture thresholding using an iterative selection method.608
Systems, Man and Cybernetics, IEEE Transactions on 8 (8), 630 –632.609
Rigby, S. P., Chigada, P. I., Wang, J., Wilkinson, S. K., Bateman, H., Al-Duri, B., Wood, J.,610
Bakalis, S., Miri, T., 2011. Improving the interpretation of mercury porosimetry data using611
computerised x-ray tomography and mean-field dft. Chemical Engineering Science 66 (11),612
2328 – 2339.613
Rondeau, X., Affolter, C., Komunjer, L., Clausse, D., Guigon, P., 2003. Experimental determi-614
nation of capillary forces by crushing strength measurements. Powder Technology 130 (1-3),615
124–131.616
Roscoe, K. H., 1970. The influence of strains in soil mechanics. Geotechnique 20 (2), 129–170.617
Russell, A. R., 2010. Water retention characteristics of soils with double porosity. European618
Journal of Soil Science 61 (3), 412–424.619
Tarantino, A., De Col, E., 2008. Compaction behaviour of clay. Geotechnique 58 (3), 199–213.620
Taud, H., Martinez-Angeles, R., Parrot, J., Hernandez-Escobedo, L., 2005. Porosity estimation621
method by x-ray computed tomography. Journal of Petroleum Science and Engineering 47 (3–622
4), 209–217.623
Thevanayagam, S., Nasarajah, S., 1998. Fractal model for flow through saturated soils. Journal624
of Geotechnical and Geoenvironmental Engineering 124 (1), 53–66.625
Tolle, C. R., Mcjunkin, T. R., J., G. D., 2003. Suboptimal minimum cluster volume cover-626
based method for measuring fractal dimension. In: IEEE Transaction on Pattern Analysis627
and Machine Intelligence. pp. 32–41.628
21
Tuller, M., Or, D., Dudley, L. M., 1999. Adsorption and capillary condensation in porous media:629
Liquid retention and interfacial configurations in angular pores. Water Resour. Res. 35 (7),630
1949–1964.631
Vallejo, L. E., 1996. Fractal analysis of the fabric changes in a consolidating clay. Engineering632
Geology 43 (4), 281–290.633
Vallejo, L. E., 2010. Compressive strength of unsaturtated fissured clays. In: Alonso, E., Gens,634
A. (Eds.), Unsaturated soils. Fifth International Conference on Unsaturated Soils, CRC Press,635
pp. 387–390.636
Van Geet, M., Swennen, R., Wevers, M., 2000. Quantitative analysis of reservoir rocks by637
microfocus x-ray computerised tomography. Sedimentary Geology 132 (1-2), 25–36.638
Van Geet, M., Volckaert, G., Roels, S., 2005. The use of microfocus x-ray computed tomography639
in characterising the hydration of a clay pellet/powder mixture. Applied Clay Science 29 (2),640
73–87.641
Wellington, S. L., Vinegar, H. J., 1987. X-ray computerized tomography. Journal of Petroleum642
Technology, 885–898.643
Xu, Y. F., 2004a. Calculation of unsaturated hydraulic conductivity using a fractal model for644
the pore-size distribution. Computers and Geotechnics 31 (7), 549–557.645
Xu, Y. F., 2004b. Fractal approach to unsaturated shear strength. Journal of Geotechnical and646
Geoenvironmental Engineering 130 (3), 264–273.647
Xu, Y. F., Sun, D. A., 2002. A fractal model for soil pores and its application to determination648
of water permeability. Physica a-Statistical Mechanics and Its Applications 316 (1-4), 56–64.649
Zeng, Y., Gantzer, C. J., Payton, R. L., Anderson, S. H., 1996. Fractal dimension and lacunarity650
of bulk density determined with X-ray computed tomography. Soil Science Society of America651
Journal 60 (6), 1718–1724.652
Zhang, L. M., Li, X., 2010. Microporosity structure of coarse granular soils. J. Geotech. and653
Geoenvir. Engrg. 136 (10), 1425–1436.654
22
first step second step third step
λ1λ2
λ3
Figure 1: Box-counting method for determining particle profile fractal dimension. Hatchingindicates those boxes that contain segments of the particle profile.
topview
sectionview
mould double-layer press single-layer press
40 mm
41mm
OD
15mm
ID
(steel) (steel) (steel)
30 mm 20 mm
Figure 2: Sketch of the mould (left) and presses (middle and right) used for specimen preparation
23
(a) (b)
≈10.5≈10.5
≈20.0
≈10.5
≈10.5
≈20.0
Figure 3: Reconstructed 3-D pore network for single-layer specimens as detected using XRCTIsoData thresholding: a) specimen 2s; b) specimen 10s
(a) (b)
≈10.5
≈10.5
≈7.5
≈10.5≈10.5
≈7.5
Figure 4: Reconstructed 3-D pore network for single-layer specimens as detected using XRCTIsoData thresholding: a) top layer of specimen 13d; b) top layer of specimen 15d
24
X
Y
Z
(a) (b) (c)
image slice
15.0mm ≈10.3mm (938 pixels)
≈9.4m
m(856
pixels)
Figure 5: Example of XRCT images: a) specimen axes; b) XY-slice pre-processing; c) XY-slicepost-processing. Outlines of particles seen in (b) have been outlined in (c).
25
import image sequence
analyse set scale 1 pixel = 11 microns
image type 8 bit (needed for thresholding to work)
edit selection specify (938 x 856 pixels, manually centre, check to see if all images OK)
image adjust
delete first and last 100 images to prevent shadowing
process filters median (2 pixels)
image adjust brightness/contrast (change min/max values to bracket intensity range)
image adjust threshold (isodata, auto)
process remove outliers (2.0, 50, dark)
process binary fill holes
— if necessary —
analyse set measurements (choose appropriate parameters)
analyse analyse particles (1089-infinity, 0.00–1.00 circularity, display results, summarise)
analysis stages
pore area data
fractal data:
plugins fractal FracLac 2.5 release 1e
standard box count grid positions: 5
binary or grayscale: use binary
background colour: let program decide
type of series: use default box sizes
sizes per series: 0
minimum size: 1
maximum box size: 45% ROI
remaining options: ALL deselected
analysis
brightness/contrast (change min/max values to bracket intensity range)
thresholding process
noise
find the minimum cover
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
16)
15)
scan image stack
or ROI
image crop
14)
file image sequencesave as13)
Figure 6: Flowchart of image processing algorithm (using ImageJ software)
26
0 5 10 15 200
5
10
15
20
25
Depth (mm)
Por
e A
rea
Fra
ctio
n (%
)
Sample 10s (114)Sample 10s (75)Sample 10s (50)Sample 2s (107)Sample 2s (75)Sample 2s (50)
Figure 7: Change in calculated PAF with depth for different threshold intensity values (givenin parentheses)
0 5 10 15 20
0.8
1
1.2
1.4
1.6
1.8
2
Depth (mm)
Min
imum
Cov
er F
ract
al D
imen
sion
, Df
Sample 10s (114)Sample 10s (75)Sample 10s (50)Sample 2s (107)Sample 2s (75)Sample 2s (50)
Figure 8: Change in calculated Df with depth for different threshold intensity values (given inparentheses)
27
0 5 10 15 200
0.5
1
1.5
2
2.5
3
Depth (mm)
Lacu
narit
y, L
Sample 10s (114)Sample 10s (75)Sample 10s (50)Sample 2s (107)Sample 2s (75)Sample 2s (50)
Figure 9: Change in calculated L with depth for different threshold intensity values (given inparentheses)
15.0mm 15.0mm
(a) (b)
Figure 10: XRCT images taken at: a) 16500 µm; and b) 17379 µm
28
≈10.3mm
≈9.4m
m
(a) (b) (c) (d)
Figure 11: Large particle present in specimen 10s at 13090 µm: a) XRCT image; b) thresholdintensity of 50; c) threshold intensity of 75; d) threshold intensity of 114 (IsoData value).
0 5 10 15 20-80
-60
-40
-20
0
20
40
60
80
100
Depth (mm)
Rel
ativ
e P
AF
(%
)
2s & 10s (Auto)2s & 10s (75)2s & 10s (50)
Figure 12: Change in calculated relative PAF with depth for different threshold intensity values
29
0 5 10 15 20-10
-5
0
5
10
15
20
25
30
35
Depth (mm)
Rel
ativ
e M
inim
um C
over
Fra
ctal
Dim
ensi
on (
%)
2s & 10s (Auto)2s & 10s (75)2s & 10s (50)
Figure 13: Change in calculated relative Df with depth for different threshold intensity values
0 5 10 15 20-250
-200
-150
-100
-50
0
50
Depth (mm)
Rel
ativ
e La
cuna
rity
(%)
2s & 10s (Auto)2s & 10s (75)2s & 10s (50)
Figure 14: Change in calculated relative L with depth for different threshold intensity values
30
0 5 10 15 2010
-1
100
101
102
Depth (mm)
Por
e A
rea
Fra
ctio
n (%
)
Sample 13d (141)Sample 13d (60)Sample 15d (122)Sample 15d (60)
Figure 15: Change in pore area fraction with depth for specimens 13d and 15d for differentthreshold intensity values
0 5 10 15 20
0.8
1
1.2
1.4
1.6
1.8
2
Depth (mm)
Min
imum
Cov
er F
ract
al D
imen
sion
, Df
Sample 13d (141)Sample 13d (60)Sample 15d (122)Sample 15d (60)
Figure 16: Change in minimum cover fractal dimension with depth for specimens 13d and 15dfor different threshold intensity values
31
0 5 10 15 200
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Depth (mm)
Lacu
narit
y, L
Sample 13d (141)Sample 13d (60)Sample 15d (122)Sample 15d (60)
Figure 17: Change in lacunarity with depth for specimens 13d and 15d for different thresholdintensity values
32
(a) (b) (c)
(d) (e) (f)
≈10.3mm
≈9.4m
m
Figure 18: Effect of changing the threshold value on the resulting binary images for specimen13d: a) original image; b) analysed using threshold value of 60; c) analysed using IsoDatathreshold value; d) original image; e) analysed using threshold value of 60; f) analysed usingIsoData threshold value.
33
20.0mm
15.0mm
toplayer
bottom
layer
Figure 19: XZ-plane central section of specimen 13d
34
Table 1: OWC, ρdmax and ρbmax for RE soil mix
OWC (%) ρdmax (kg/m3) ρbmax (kg/m3)
12.0 1918.1 2147.0
Table 2: Specimen preparation
Specimen Layers Applied load (% of UCS) IsoData Thresholding value (intensity)
2s 1 85 10710s 1 25 11413d 2 0 14115d 2 0 122
35
Top Related